The Tables 7.1 and 7.2 provide an overview of the sensitivity study results described above. Listed in Table 7.1 are the retrieval errors – at the five different latitudes – associated with tropospheric clouds and with uncertainties in the knowledge of surface albedo and neutral density profile. Table 7.2 shows the retrieval errors corresponding to uncertainties in the ozone profile for a polar (83◦N) and a tropical (0◦) viewing geometry, and in tangent height registration at 40◦N. The assumed albedo error is 0.15 (with respect to A = 0.5), the assumed error in the neutral density profile is 3 % (scaling with altitude independent factor), and the error in the ozone profile is 15 % (scaling with altitude independent factor). The tangent height error is 200 m. To determine the error with respect to clouds, a typical tropospheric cloud (following Sonkaew et al. (2009) and references therein) is simulated, i.e., a water cloud with a droplet radius of 8 µm, an optical thickness of 10 and a vertical extension from 4–7 km. Again, the ground albedo is 0.3.
7.1 HENYEY-GREENSTEIN APPROXIMATION 113
Alt. [km] 83◦N 40◦N 0◦ 40◦S 75◦S
A ND Cl. A ND Cl. A ND Cl. A ND Cl. A ND Cl.
15 <1 5 1 1 8 1 5 10 1 8 12 4 1 20 15
20 <1 2 2 2 1 2 7 2 2 12 3 12 3 15 25
25 <1 0 2 2 1 3 8 2 3 15 1 17 3 5 30
30 <1 2 2 2 2 3 8 2 3 15 0 15 4 5 25
35 <1 2 1 3 2 2 8 3 2 15 2 1 5 5 10
Mean < 1 2 2 2 3 2 7 4 2 13 5 10 3 10 20
Table 7.1: Absolute value of the relative error [%] of the aerosol extinction retrieved with the Henyey-Greenstein retrieval version (V1.0) due to uncertainties in ground albedo (“A”, uncertainty 0.15) and neutral density (“ND”, 3 %) and due to a typical tropospheric cloud (“Cl.”) for the SCIAMACHY viewing geometry at 83◦N, 40◦N, 0◦, 40◦S, and 75◦S.
The results show that retrieval version V1.0 using a Henyey-Greenstein phase function is very sensitive to errors in other geophysical parameters, particularly in the southern hemisphere. Tropospheric clouds and uncertainties in the knowledge of neutral density and ground albedo can lead to errors of up to 30 %, 20 %, and 15 %, respectively. Smaller error sources are tangent height registration (8 %), the a
priori profile (7 %), and ozone (2 %).
114 7 SENSITIVITY STUDIES
Alt.[km] O3 TH 83◦N 0◦ 40◦N
15 2 1 8
20 2 1 2
25 1 0.5 3
30 0 0.5 5
35 0 0.5 3
Mean 1 1 4
Table 7.2: Absolute value of the relative error [%] of the aerosol extinction retrieved with the Henyey-Greenstein retrieval (V1.0) version due to uncertainties in the ozone profile (15 %) and tangent height (200 m) for the SCIAMACHY viewing geometry at 83◦N and 0◦ and 40◦N, respectively.
7.2 MIE PHASE FUNCTION 115
7.2 Mie phase function
The sensitivity studies for the HG retrieval version showed a strong dependence on the latitude of the SCIAMACHY observations, which is a consequence of the aerosol scattering phase function in combination of the latitudinal dependence of the scattering angle. The following Sections 7.2.1 – 7.2.6 show the results of the same studies using the Mie phase function described in Section 6.1.2 and point out the differences to the Henyey-Greenstein retrieval version. In Section 7.2.7, the results are summarized. Unless mentioned otherwise in the text, all parameters except the phase function remained unchanged for a better comparison with the HG version.
7.2.1 Impact of a priori profile
Figure 7.8 shows the results of the synthetic retrievals with the six different true aerosol extinction profiles generated by modifications of the a priori profile described in Section 7.1.1. A different a priori profile than for the HG version is used, i.e. the one that is used for the retrieval of real SCIAMACHY data later in this work (the light blue curve for 75◦S in Figure 6.4). The modifications are the same as for the HG version. The results are very similar to those of the HG version, the relative error is below 10 % for all cases and altitudes. The only mentionable difference appears in altitudes between 30 and 40 kilometers for the± 3 km shift (panels c and d of Figure 7.8), were the relative error is larger than for the HG phase function, but still relatively small (within± 7 %).
116 7 SENSITIVITY STUDIES
Figure 7.8: As Figure 7.1, but with the Mie phase function described in Section 6.1.2 instead of the Henyey-Greenstein approximation. Left panels: true and a priori aerosol extinction profiles. Right panels: relative difference between the true and the retrieved aerosol extinction profiles. Modifications of the ”true” profile in comparison to the a priori profile: a) multiplied by 0.5; b) multiplied by 2; c) altitude shift + 3 km; d) altitude shift
−3 km; e) artificial maximum around 25 km; f) artificial minimum around 25 km.
7.2 MIE PHASE FUNCTION 117
7.2.2 Effect of surface albedo
Equivalent to Figure 7.2, Figure 7.9 shows the relative error in the retrieved aerosol extinction coefficient for true albedo values between 0 and 1 with respect to an assumed albedo of 0.5 for the five SCIAMACHY geometries. The two main observa-tions described in Section 7.1.2 – a much higher error in the tropics compared to the polar regions and a much higher sensitivity to the ground albedo in the southern hemisphere – are still visible, but both less pronounced. The reason for this is the reduced sensitivity at the equator and at 40◦S, where the absolute value of the largest relative error is reduced by ≈20 %. This is a result of the higher value of the Mie phase function compared to the Henyey-Greenstein approximated one at scattering angles corresponding to these latitudes, which means that less aerosol extinction is needed to compensate for the same error. The changes in the polar regions are small compared to the HG version, the already small relative error at 75◦S is even smaller for the Mie phase function. Note again the feature below 20 km at this latitude, a result of the very large solar zenith angle. The sign of the relative error switches between 40◦N and the equator, while in the HG version this happens between 83◦N and 40◦N. This is also a result of the different behaviour in the forward scattering area of the new phase function (see Figure 6.1) that corresponds to single scattering for SCIAMACHY geometries in the northern hemisphere (see Table 5.1). At 40◦N, an overestimation or underestimation of the ground albedo would now lead to an aerosol extinction error with different sign than for the HG case, but with relatively small absolute values of generally smaller than 10 % in both cases.
118 7 SENSITIVITY STUDIES
Figure 7.9: As Figure 7.2, but with the Mie phase function described in Section 6.1.2 instead of the Henyey-Greenstein approximation: relative errors in the aerosol extinction coefficients for different values of the true surface albedo A and the SCIAMACHY viewing geometry at 83◦N, 40◦N, 0◦, 40◦S, and 75◦S. The albedo assumed for the retrieval is A
= 0.5.
7.2 MIE PHASE FUNCTION 119
7.2.3 Effect of neutral density
In Figure 7.10 the relative errors in the retrieved aerosol extinction for artificial errors of± 3 % in the neutral density – simulated by corresponding changes of the ground pressure, see Section 7.1.3 – are depicted. Compared to the HG version, the errors are similar for 83◦N, 40◦N, and the equator, and smaller for the two southern latitudes. The largest effect can be seen in altitudes above 35 km at 40◦S and below 35 km at 75◦S. Above 20 km the error is now≤5 % for all latitudes.
7.2.4 Effect of ozone
Figure 7.11 shows the result for the ozone sensitivity study for the retrieval with a Mie phase function. For the reasons mentioned in Section 7.1.4, the impact of errors in the ozone profile on the retrieved aerosol extinction is negligible,<2 % even for the polar geometry with a high ozone column.
7.2.5 Effect of tropospheric clouds
Figure 7.12 shows the result for the cloud sensitivity study for the retrieval with a Mie phase function. As for the previous parameters, the error in the aerosol extinction is significantly reduced after implementing a Mie phase function. Even for this optically thick cloud layer the relative error at all latitudes except 75◦S is rather small,<15 % at all altitudes. The effect of the large solar zenith angle at 75◦S is still visible, but not as pronounced as for the Henyey-Greenstein version. A low cloud (1–4 km, red curve) now leads to an overestimation in the aerosol extinction of up to 40 % from above the tropopause to 32 km and an underestimation of up to 40 % above, a high cloud (6–9 km, green curve) to a strong underestimation (up to 60 %) in the aerosol extinction below 16 km and an overestimation of up to 50 % above. (Note that the highest cloud with a CTH of 10 km leads to a failure of the
120 7 SENSITIVITY STUDIES
Figure 7.10: As Figure 7.3, but with the Mie phase function described in Section 6.1.2 instead of the Henyey-Greenstein approximation: relative errors in the aerosol extinction coefficients for an error of about ±3 % in the neutral density and the SCIAMACHY viewing geometry at 83◦N, 40◦N, 0◦, 40◦S, and 75◦S.
7.2 MIE PHASE FUNCTION 121
Figure 7.11: As Figure 7.4, but with the Mie phase function described in Section 6.1.2 instead of the Henyey-Greenstein approximation: relative errors in the aerosol extinction coefficients for an error of±15 % in the ozone profile for a polar and a tropical SCIAMACHY viewing geometry. Left panel: 83◦N, 400 DU±15 %; right panel: 0◦, 250 DU±15 %.
retrieval.) After reducing the SZA to 83◦(see last panel in Figure 7.12) the relative error in the aerosol extinction is with±20 % still relatively large compared to the other latitudes, but much smaller than in retrieval version V1.0.
The comparison with an ice cloud (left set of panels in Figure 7.13) and an optically thinner cloud (τ = 10, right set of panels) at 40◦S leads to the same conclusion as for retrieval version V1.0: the aggregation state only plays a minor role with respect to influencing the retrieved aerosol extinction, a smaller optical thickness leads to a smaller error in the aerosol extinction.
7.2.6 Effect of tangent height errors
The results of repeating the sensitivity study for tangent height errors for the Mie phase function are shown in Figure 7.14. The sensitivity of V1.1 is smaller compared to V1.0, the error for the± 1000 m shift is <20 % above 20 km and <35 % below.
For a typical SCIAMACHY tangent height error of± 200 m, the largest impact on the aerosol extinction is about 6 % at 16 km and<4 % above 20 km.
122 7 SENSITIVITY STUDIES
Figure 7.12: As Figure 7.2, but with the Mie phase function described in Section 6.1.2 instead of the Henyey-Greenstein approximation: Relative error in the aerosol extinction (V1.1) for a water cloud layer with 3 km vertical extension and an optical thickness ofτ
= 20 at seven different altitudes below the retrieval altitude. The first five sets of panels again show aerosol extinction profiles (left panels) and the relative error (right panels) for a SCIAMACHY viewing geometry at 83◦N, 40◦N, 0◦, 40◦S, and 75◦S as listed in Table 5.1. The last set of panels (bottom right) shows the 75◦S geometry with a modified solar zenith angle of 83◦S.
7.2 MIE PHASE FUNCTION 123
Figure 7.13: Relative aerosol extinction retrieval error for a cloud layer with 3 km vertical extension at seven different altitudes below the retrieval altitude for a SCIAMACHY viewing geometry at 40◦S. Left panel: ice cloud withτ = 20. Right panel: water cloud with τ = 10.
Figure 7.14: As Figure 7.7, but with the Mie phase function described in Section 6.1.2 instead of the Henyey-Greenstein approximation: relative errors in the aerosol extinction coefficients for a tangent height error of±200 m, ±500 m, and ±1000 m for a northern mid-latitude SCIAMACHY viewing geometry.
124 7 SENSITIVITY STUDIES
7.2.7 Summary of the sources of potential systematic errors
Similar to the Tables 7.1 and 7.2 for the retrieval with a Henyey-Greenstein approxi-mation, the Tables 7.3 and 7.4 summarize the results of the sensitivity studies for the retrieval with a Mie phase function. In general, the retrieval with an implemented Mie phase function is much less sensitive to errors in other geophysical parame-ters than the Henyey-Greenstein version. Especially in the southern hemisphere the effect is apparent, the error induced by the same uncertainty in a geophysical parameter is reduced by a factor 2. As a result, the difference in sensitivity between the two hemispheres – visible most clearly comparing Figure 7.2 and Figure 7.9 – is reduced significantly.
The largest source for errors are uncertainties in the knowledge of neutral den-sity (15 % error at most), tropospheric clouds (9 %) and ground albedo (8 %) in the southern hemisphere, followed by the a priori profile (7 %), tangent height registration (6 %), and ozone (<1 %).
Alt. [km] 83◦N 40◦N 0◦ 40◦S 75◦S
Table 7.3: As Table 7.1, but with a Mie phase function instead of the Henyey-Greenstein approximation: absolute value of the relative error [%] of the aerosol extinction (V1.1) due to uncertainties in ground albedo (“A”, uncertainty 0.15) and neutral density (“ND”, 3 %) and due to a typical tropospheric cloud (“Cl.”) for the SCIAMACHY viewing geometry at 83◦N, 40◦N, 0◦, 40◦S, and 75◦S.
7.2 MIE PHASE FUNCTION 125
Alt. [km] O3 TH
83◦N 0◦ 40◦N
15 1 0.5 6
20 1 0.5 2
25 0 0.5 2
30 0 0 3
35 0 0 2
Mean 0.5 0.5 3
Table 7.4: As Table 7.2, but with a Mie phase function instead of the Henyey-Greenstein approximation: absolute value of the relative error [%] of the aerosol extinction (V1.1) due to uncertainties in the ozone profile (15 %) and tangent height (200 m) for the SCIAMACHY viewing geometry at 83◦N and 0◦ and 40◦N, respectively.
126 7 SENSITIVITY STUDIES
Part III
Retrieval results
129
8 Comparison with SAGE II data
After testing the sensitivity of the algorithm to various parameters, we applied the retrieval approach described in Section 5.1 to the entire SCIAMACHY data set of limb-scatter observations from August 2002 – April 2012. To demonstrate the quality of the algorithm, we validated our results with co-located SAGE II measurements. As described in Section 1.1, the SAGE solar occultation instrument series and particularly SAGE II provided a stratospheric aerosol extinction data set which is widely regarded as highly accurate and therefore well suited for the validation of the SCIAMACHY results presented here. One of the data products of SAGE II are aerosol extinction profiles at 525 nm wavelength. We converted the SCIAMACHY stratospheric aerosol extinction values to this wavelength using the assumed spectral dependence of the aerosol extinction coefficient with Ångstrøm exponents delivered by SCIATRAN, approximately 1.54 for V1.0 and 1.43 for V1.1.
The 525 nm profiles were then compared to co-located SAGE II profiles within a spatial distance of 500 km and a temporal difference of six hours at most during the temporal overlap of the two missions between 01 January 2003 (start of the SCIAMACHY routine operations phase) and 17 August 2005 (demise of SAGE II).
SCIAMACHY data with a SZA exceeding 87◦were not considered in this comparison.
Recently a new version (V7.0) of the SAGE II aerosol extinction data set has been published (Damadeo et al., 2013). Changes in the algorithm and meteorological data used to derive the aerosol extinction from the occultation measurements led to significant differences in the 525 nm aerosol extinction compared to V6.2 published
130 8 COMPARISON WITH SAGE II DATA
in 2003, particularly above 20 km altitude. In this chapter both SCIAMACHY versions V1.0 and V1.1 are compared with both SAGE II versions V6.2 and V7.0.
The first part of this chapter deals with retrieval version V1.0 containing the Henyey-Greenstein phase function. In the second part, the same procedure is repeated with the Mie phase function (retrieval version V1.1) and the results are compared to V1.0. To investigate the effect of clouds on the retrieval, the comparisons of both retrieval versions are additionally shown in the corresponding section using cloud occurence information that is also obtainable from SCIAMACHY radiance spectra.
8.1 Henyey-Greenstein approximation
Figure 8.1 shows the comparison – mean profiles and relative differences (SCIA-MACHY−SAGE)/SAGE – of the globally averaged SAGE II V6.2 (black) and V7.0 (blue) with SCIAMACHY stratospheric aerosol extinction profiles V1.0 (red) at 525 nm wavelength. Below 15 km, SCIAMACHY overestimates the aerosol extinc-tion by up to 40 % as compared to both versions of SAGE II. But above 15 km the agreement to V6.2 is within 20 % and between 16 and 30 km even within about 10 %. Compared to V7.0, SCIAMACHY overestimates the stratospheric aerosol
extinction by approximately 20 % between 15 and 32 km.
To gain more information about the meridional behavior of the retrieval, the profiles were averaged zonally and over all available co-locations as well as binned into eight 20◦latitude bins between 80◦N and 80◦S (Figure 8.2). Table 8.1 shows the relative differences of the retrieved aerosol extinction in comparison to co-located SAGE II measurements for the eight latitude bins. Between 20◦N and 20◦S values for 15 km altitude have been ignored because of tropospheric influences at these latitudes. For the same reason, the 15 km values for 20–40◦ N/S are put in
8.1 HENYEY-GREENSTEIN APPROXIMATION 131
Figure 8.1: Left panel: comparison of average co-located SAGE II V6.2 (black)/V7.0 (blue) and SCIAMACHY (red) 525 nm aerosol extinction profiles (V1.0) with standard deviation (dashed lines). The green line shows the a priori extinction profile used for the SCIAMACHY retrievals. The number in the top right corner shows the number of co-locations averaged. Right panel: mean relative difference between SCIAMACHY and both versions of SAGE II aerosol extinction profiles with standard deviation (dashed).
parentheses. In the following, first the comparison to V6.2 (black curves in Figure 8.2) is discussed.
At low latitudes (20◦N–20◦S) the agreement is quite good above 20 km altitude.
Above the tropopause the relative difference is generally smaller than 20 % in both hemispheres. Above 20 km altitude the SCIAMACHY aerosol extinctions in the southern hemisphere are up to 20 % larger than the SAGE II values. Below 20 km the frequent occurrence of tropospheric clouds makes SCIAMACHY and SAGE II products uncomparable. An analysis of the tropospheric cloud detection data set obtained with SCODA (SCIAMACHY ClOud Detection Algorithm) (Eichmann et al., 2009) showed that about 95 % of all SCIAMACHY limb measurements are affected by tropospheric clouds.
132 8 COMPARISON WITH SAGE II DATA
Figure 8.2: Left panels: comparison of the retrieved (V1.0) 525 nm aerosol extinction profiles (red) with SAGE II V6.2 (black) and V7.0 (blue) aerosol extinction in eight latitude bins with standard deviation (dashed lines). The a priori profile is shown in green.
The numbers in the top right corner show the number of co-locations averaged (“#”) and the average scattering angle (“S”). Right panels: mean relative difference between SCIAMACHY and both versions of SAGE II aerosol extinction profiles with standard deviation (dashed).
8.1 HENYEY-GREENSTEIN APPROXIMATION 133
Table 8.1: Relative difference [%] of the SCIAMACHY aerosol extinction in comparison to co-located SAGE II V6.2 and V7.0 measurements, both averaged in eight latitude bins. The Henyey-Greenstein approximation for the phase function was used (V1.0).
Between 20◦N and 40◦N, the relative differences are generally within ± 20 %.
Only above 30 km and below 17 km the differences are larger. In the corresponding region in the southern hemisphere, the shape of the relative difference profile is similar, but the values are larger than in the northern hemisphere, ranging from
−50 % at 35 km altitude to +100 % near 15 km. However, between 20 and 30 km, the difference is generally within± 20 %. The large error below 20 km might still be an effect of tropospheric clouds.
Between 40◦N and 60◦N, the agreement is quite good below 25 km with relative differences to SAGE of ± 20 % at most. Above 25 km, SCIAMACHY results are systematically lower than those of SAGE II, e. g., by about 50 % at 32 km. The picture of the corresponding latitude bin in the southern hemisphere looks completely different. At all altitudes below 30 km, we see systematically higher values. At 20 km the relative difference reaches 60 %.
For latitudes between 60◦ and 80◦ we observe a significant interhemispheric difference in the relative differences between SCIAMACHY and SAGE II aerosol extinction profiles. In the northern hemisphere the relative differences are negative for all altitudes between 15 and 34 km (with a maximum difference of about−50 % between 25 and 30 km), whereas they are positive for all altitudes in the southern
134 8 COMPARISON WITH SAGE II DATA
hemisphere remaining more or less constant at about+30 % between 15 and 30 km.
Summarized, the good globally averaged agreement of SCIAMACHY V1.0 with SAGE II V6.2 gets bad on closer inspection of the latitudinal behavior.
The comparison with SAGE II V7.0 (blue curves in Figure 8.2) looks slightly different, but not better. The difference at almost all latitudes and altitudes is shifted towards more positive values by approximately 5–10 %. This leads to worse agreement in all latitude bins except the ones representing mid and high northern latitudes (40◦–80◦N). The overall picture and therefore the following interpretation is valid for both SAGE II versions.
In general it can be said that the SCIAMACHY retrieval results in lower strato-spheric aerosol extinction in the northern hemisphere and higher values in the southern hemisphere as compared to SAGE II measurements (see Figure 8.3). One possible reason for this apparent interhemispheric difference is the different sensitiv-ity of the aerosol profile retrievals to errors in the surface albedo (see Section 7.1.2, in particular the comparison of 40◦N and 40◦S in Figure 7.2). Tropospheric clouds affect the majority of the SCIAMACHY limb observations, and this will lead to a higher effective albedo at 470 nm and 750 nm as compared to the surface albedo data base by Matthews (1983) currently used by SCIATRAN for the retrieval. This is particularly true for measurements above the ocean. A low bias in the assumed albedo is most commonly associated with high bias in the aerosol extinction values, and this effect is much stronger in the southern hemisphere. This is qualitatively con-sistent with the validation results presented in Figure 8.2 – particularly considering the comparisons for 40◦–60◦N/S and 60◦–80◦N/S.
This aspect was investigated further by applying the SCODA cloud detection data
This aspect was investigated further by applying the SCODA cloud detection data